A multi-agent approach to distributed control for task-level programs for cooperating manipulators

The use of two or more robots in a common workspace is essential to expand the field of potential applications. This paper presents a distributed approach for executing task-level programs for cooperating manipulators or one arm systems with many degrees of freedom. At task level, only the movement of the objects to be manipulated is specified. The according manipulator motions are then calculated by assigning local intelligence to each joint. The local intelligence enables the joint agents to calculate in parallel their appropriate movements to make the end effector reach the desired position. During execution, each joint evaluates sensor data to compensate for execution errors, to react on unexpected obstacles and to provide the manipulator coordination. Therefore, a distributed fuzzy rule base has been developped. As this local evaluation may lead to non-optimal overall behavior of the manipulator, the joint agents communicate to provide global suboptimality.